Tests are not fully deterministic

[JelleAalbers]

One of the builds in a metadata-only PR (#219) just failed with:

>       ll = lf.limit('er_rate_multiplier', bestfit,
                      confidence_level=0.9, kind='lower')

tests/test_inference.py:88: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
flamedisx/likelihood.py:707: in limit
    res = opt(
flamedisx/inference.py:374: in minimize
    result, llval = self.parse_result(result)
flamedisx/inference.py:461: in parse_result
    self.fail(f"Scipy optimizer failed: "
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <flamedisx.inference.ScipyIntervalObjective object at 0x7f83a450feb0>
message = 'Scipy optimizer failed: status = 0: The maximum number of function evaluations is exceeded.'

    def fail(self, message):
        if self.allow_failure:
            warnings.warn(message, OptimizerWarning)
        else:
>           raise OptimizerFailure(message)
E           flamedisx.inference.OptimizerFailure: Scipy optimizer failed: status = 0: The maximum number of function evaluations is exceeded.

flamedisx/inference.py:396: OptimizerFailure

but succeeded on a rerun. Apparently our tests are not fully deterministic.